KA seems to be broken for CUDA

Created by: benedict-96

Whenever I try to call a function that should use my GPU (GeForce RTX 4090) I either get an overflow error or a never-ending loop. For example:

dev = CUDA.device()
KernelAbstractions.zeros(dev, Float32, 10, 5) 

or

get_backend(cu(rand(1)))

.